The Multistage Support Vector Machine

Support vector machines (SVMs) for binary classification have been solved perfectly, but SVMs for multiclass classification and regressive ability need to be researched and improved further. In this dissertation, we provide how to apply support vector regression to function approximation by means of experiments and obtain the relevant conclusion. We also put forward the concept, algorithm and examples of multistage support vector machine which based on the idea of classification after clustering.While using support vector regression to do function approximation, we can control the number of support vectors and the approximative performance all by two parameters.The advantage of multistage support vector machine is embodied in three aspects. First, towards the unpredicted areas of other multiclass support vector machines, multistage support vector machine can predict them more correctly; Secondly, according to the experimental comparison, the dissertation shows us the high accuracy of its evaluate performance. Finally, for a multiclass classification, multistage support vector machine need less support vectors to construct multistage hyperplane than the other three methods, so the multistage support vector has the better generalization.